• No results found

In this section we evaluate different methods described in Table 7.3 for accuracy using different decision- making algorithms. Until now, evaluation for accuracy was using QoI-based algorithm. This is because of the weighted majority considered by QoI-based that accounts the age of the notification as well. This feature is beneficial in vehicular environments such as our scenario where temporal dynamics of an event is significant for quicker adaptability to new road conditions.

A Majority based decision-making algorithm applies simple majority based resolution. Since in our scenario the road condition is represented in a binary value either with a traffic-jam or a no-traffic-jam state, majority based decision-making can be employed to make a rerouting decision. On the contrary, with the Newest-based decision-making, a vehicle always makes rerouting decision based on the latest value in its cache. In this case, all other cache entries except recent entry are irrelevant.

Accuracy at 20% False Detection Rate:

Figure 7.14a gives an overview of accuracy achieved using different decision-making algorithms in a 20% false detection environment. Considering the dynamic nature of vehicular environment with frequent changes in traffic behaviour, it can be observed that QoI-based decision algorithm per- forms better than other algorithms. As discussed in previous evaluations using QoI-based decision- making, the methods Default(none) and Co-op.+ACM(vary.VR) performs better than the methods Co-

op.(none) and Co-op.+ACM(const.VR). It can be noted that besides poor detection accuracies, method Co-op.+ACM(vary.VR) consistently achieves accuracy almost as close as the method Default(none) which

achieves best accuracy levels due to redundant publications.

With respect to Newest-based algorithm, since decision is based only on the latest entry in cache, all the methods involving Cooperative-approach of reporting perform the same. However, there is a slight degradation in accuracy achieved by methods involving Cooperative-approach when compared with Default-approach. This slight degradation in performance can be related to the cost of unaware- vehicles. As unaware-vehicles which have not received any notification of the existing traffic-jam state, only upon their arrival at POI they decide to publish and update their cache. In comparison to QoI- based algorithm, Newest-based performs lower due to slow adaptation to changing road status in false detection environment.

In contrast to QoI-based and Newest-based algorithms, Majority-based algorithms achieve the least accuracy. It is lower than other two algorithms as neither does it regard the age of the information as QoI-based, nor does it consider the newest information which can give the recent road status. Instead, Majority-based is dependent on simple majority of values for decision-making which relies also on out- dated information. Due to this characteristic, method Co-op.(none) though without accounting vehicles which did not publish and henceforth has the least cache entries among all, performs better than the method Default(none) which actually accounts all the publishing vehicles. However, in comparison be- tween the methods Default(none) and Co-op.+ACM(vary.VR), it can be noted, they both achieve almost the same performance in Majority-based decision-making.

Accuracy at 0% False Detection Rate:

Figure 7.14b gives an overview of accuracy achieved using different decision-making algorithms in a 0% false detection environment. Without any false detections and false reporting, QoI-based and Newest- based algorithm achieve the same performance levels across all methods. This is because all the changes of road state are correctly reported without any false notifications. This is also the reason for observing no difference in performance for all the methods that follow Cooperative-approach.

It is evident that irrespective of any decision-making algorithm at any false detection rate, Co-

op.+ACM(const.VR) performs badly. This is due to considering a constant vehicular rate of 1 vehicle/sec-

ond leading to either under-estimation or over-estimation of virtual entries. Leaving out this method, in comparison between Co-op.(none) and Co-op.+ACM(vary.VR), one can choose Co-op.+ACM(vary.VR) method for higher accuracy across all algorithms except Majority-based. Though for Majority-based at

QoI-based

Majority-based

Newest-based

Decision Making

0.5

0.6

0.7

0.8

0.9

1.0

P

ercen

tage

of

correct

information

Default(none) Co-op.(none) Co-op.+ACM(const.VR) Co-op.+ACM(vary.VR)

(a)in 20% False Detection Rate Environment

QoI-based

Majority-based

Newest-based

Decision Making

0.5

0.6

0.7

0.8

0.9

1.0

1.1

P

ercen

tage

of

correct

information

Default(none) Co-op.(none) Co-op.+ACM(const.VR) Co-op.+ACM(vary.VR)

(b)in 0% False Detection Rate Environment

Figure 7.14:Accuracy achieved using different decision-making algorithms.

0% false detection rate there is significant difference in accuracy between these two methods, but it is very trivial in a 20% false detection rate. Due to this, choosing Co-op.+ACM(vary.VR) method with virtual entries over the method Co-op.(none) without virtual entries is advantageous as it gives better accuracy in other decision algorithms across low and high false detection rates.

Choosing between the methods Default(none) and Co-op.+ACM(vary.VR) is a trade off decision. This is because both these methods achieve the best possible accuracy across any false detection rates over other methods. Only that Default(none) performs slightly better than Co-op.+ACM(vary.VR). However, in comparison to the load exerted on broker by Default-approach and Cooperative-approach, the method

Co-op.+ACM(vary.VR) is clearly most suitable in high density vehicular networks that keeps transmission

costs minimal.

Given the significantly low number of cellular upload and download costs associated with Cooperative- approach (Figure 7.5) and the accuracy levels achieved with Co-op.+ACM(vary.VR) across varying false detection rates, which is just as good as Default(none), the method Co-op.+ACM(vary.VR) is clearly best suited for high density vehicular environments where road conditions change often.

False decisions in our study happens due to two reasons. One is due to false reporting by vehicles which detect a road situation wrongly. Another reason is due to slow adaptations to new road states. In our simulation results, it is observed that Newest-based decision algorithm performs well whenever a change

in road condition occurs. However, the same algorithm does poorly to recover from a wrong notification in high false detection rates. Therefore, Newest-based algorithm adapts quickly to changes but is not robust to false detections, and accordingly is best suited in an environment where all vehicles report correctly. Majortiy-based algorithm adapts slowly to changes and is weak to false detections as well. On the contrary, QoI-based algorithm, adapts quickly to changes as well as recovers fast besides wrong notifications. Since vehicular environment is characterized by frequent changes in road conditions and can include falsely reporting vehicles, QoI-based algorithm is reliable and preferred for decision-making.